Fei-Fei Li taught AI to see, and now AMD is buying the worlds she builds

Fei-Fei Li

Photo: ITU Pictures / CC BY 2.0, via Wikimedia Commons

Born	July 3, 1976 Beijing
Occupation	Computer Scientist
Awards	J. K. Aggarwal Prize · Queen Elizabeth Prize for Engineering · ACM Fellow

For most of her career Fei-Fei Li has warned that artificial intelligence was drifting toward the few companies rich enough to own the computers it runs on, and she spent years campaigning for public computing power so that universities could keep up. Now the startup she founded to chase her next idea is about to disappear into one of the largest chipmakers in the world. AMD has agreed to buy World Labs in an all-stock deal valued at roughly eight billion dollars, and Li will join the company as its chief scientist. The researcher who gave machines a way to see is betting that the next step needs a chip roadmap behind it.

She came to that argument from an unlikely place. Li was born in Beijing and grew up in Chengdu, in Sichuan, and was a teenager when she and her mother followed her father to Parsippany, New Jersey. She worked weekends in the family’s dry-cleaning shop through high school, and kept going home to run it while studying physics at Princeton, with dishwashing shifts on the side. That double life, the shopkeeper’s daughter and the physics student, became the spine of her memoir, The Worlds I See.

After graduating in 1999 she went to the California Institute of Technology for a doctorate in electrical engineering, completed in 2005 under Pietro Perona and Christof Koch. Her thesis studied how people take in a scene at a glance, which led her to the problem that would define her work: machines could not recognise objects because nobody had shown them enough of the world.

The answer was ImageNet. In 2007, as a young assistant professor at Princeton, Li began building a picture database far larger than most colleagues thought sensible, guided by the psychologist Irving Biederman’s estimate that people recognise around 30,000 kinds of object. With workers recruited through Amazon Mechanical Turk, her team labelled more than 14 million images across some 22,000 categories. From 2010 an annual ImageNet challenge gave researchers a common test, and in 2012 a neural network built by Geoffrey Hinton’s group beat every rival. The deep-learning boom started there, and Li’s data was the ground it ran on.

She moved to Stanford in 2009, ran its Artificial Intelligence Laboratory from 2013 to 2018 and built CS231n, a course on deep learning for computer vision whose lectures, posted online, trained a generation of engineers. In 2017 she co-founded AI4ALL, a nonprofit that runs summer programmes to bring girls and under-represented students into the field. The same year she took a sabbatical to become chief scientist for AI and machine learning at Google Cloud, where her team built tools such as AutoML.

The Google years left the most uncomfortable episode in her public record. In 2018 leaked internal emails showed Li backing Google Cloud’s part in Project Maven, a Defense Department contract to analyse drone footage, while advising colleagues to avoid any mention of its artificial-intelligence component because weaponised AI was such a sensitive subject. After employee protests Google said it would not renew the contract, and Li told The New York Times that working on any project to weaponise AI was deeply against her principles. ImageNet had its own reckoning. In 2019, after the artist Trevor Paglen and the researcher Kate Crawford built ImageNet Roulette and exposed the offensive labels sitting in its person categories, the ImageNet team said it would remove some 600,000 images of people. The dataset that taught machines to see had also taught them some of the worst ways people see each other.

Back at Stanford she co-founded the Institute for Human-Centered Artificial Intelligence in 2019 with John Etchemendy and became the best-known advocate of a national research cloud, a public pool of computing power that would let academic labs compete with industry. She sat on the federal task force that drew up its plan. Her record on regulation is less tidy than the slogan. In August 2024 she wrote in Fortune that SB 1047, a California bill holding developers liable for catastrophic misuse of their models, would shackle open-source work and cripple academic research. A month later World Labs came out of stealth with $230 million from investors led by Andreessen Horowitz, one of the venture firms that had campaigned against the bill. After the bill was vetoed she co-led the state’s expert working group, whose final report in June 2025 asked for transparency from frontier labs, independent evaluation and protection for whistleblowers.

World Labs, founded in 2024 with Justin Johnson, Christoph Lassner and Ben Mildenhall, set out to build what Li calls spatial intelligence: models that understand three-dimensional space well enough to generate it, reason about it and eventually act in it. Her case is that general intelligence needs a grounding in physics and the ability to reason about far more than text. The first product, Marble, went on general sale in November 2025 and turns a prompt, a photo or a video into an editable 3D world, pitched for entertainment and for training robots in simulation. In early 2026 the company raised another $1 billion from backers including AMD, Nvidia and Autodesk, which put in $200 million.

AMD announced the acquisition on 28 September 2026. The transaction values World Labs at about $8.2 billion, and once it closes Li becomes executive vice president and chief scientist, reporting to Lisa Su, AMD’s chair and chief executive. “Now that we have tangible proof of the possibilities, we want to do everything we can to accelerate the future,” Li wrote when the deal was announced, adding that this required getting closer to the hardware. The Bloomberg Billionaires Index estimates her fortune will reach at least $1.6 billion when the sale is completed.

Li is married to the Stanford computer scientist Silvio Savarese, and they have a son and a daughter. Time included her among the architects of AI in its 2025 Person of the Year issue, and the same year she shared the Queen Elizabeth Prize for Engineering with six other pioneers of the field, Geoffrey Hinton and Yann LeCun among them. She has said she would not call herself the godmother of AI, though she accepts the label as a way of recognising women’s work in the field.

Regulators still have to approve the sale, which AMD expects to complete before the end of 2026. After that Li will answer to a chipmaker’s product plans rather than a university calendar, and the test of her second act is concrete: whether the spatial models World Labs builds inside AMD feed the open AI ecosystem Lisa Su promised when she announced the deal, or simply become one more reason to buy AMD hardware.

Tags: business, ceos, business leaders, AMD, Fei-Fei Li, Lisa Su, Stanford University, World Labs

Related People

Related People

Discussion